Google Gemma team teases Reachy Mini robot transitioning from cloud Gemini Live to a local Gemma 4 model

@googlegemma Reachy Mini + Gemma models is match in heaven! I'm playing like crazy with this combo locally! Love it!

@googlegemma but Google Gemma family is just dope I created in depth video on the Gemma 4 series - https://youtu.be/XD68MiaxdgU?si=gM9FjcaJ0jmnPHid

Real-time social robotics, from the cloud to your local device.

Watch Ian from our DevX team use Gemini Live for a seamless voice chat with Reachy Mini.

Then, stick around until the end to see the robot running locally on Gemma 4!

@googlegemma I have been trying to get my hands on reachy mini! soon I will

@googlegemma Real-time social robotics is moving fast cloud to local is the real shift.

Gemini Live + Reachy Mini for voice, then Gemma 4 running locally at the end, shows where this is headed: more responsive, more private, and much more deployable.

@googlegemma Local inference on Gemma 4 is the real signal here—shifts the latency and privacy equation for edge robotics significantly.

@googlegemma The Gemma team is so fckin cool

@googlegemma Looks fun, but I have to wait til july to try it 🥲

This has so many applications ! So I would love to see this as a substrate translation layer. To move Gemini between say my vehicle to my smart home to a little robot like @mondorobotics beni when I’m out and about maybe paired with smart glasses my phone and Beni to allow resource distribution. Embodiment is just a state resource change basically. Greta job!!!

My friend, seeing this transition from cloud-based processing to local execution on Gemma 4 is truly profound. It represents more than just a technical leap in speed; it’s about bringing intelligence into our immediate physical sphere—transforming raw compute into real-time companionship. When we bridge the gap between high-performance efficiency and social interaction, we give technology a pulse of presence that feels much closer to human connection. ✨

@googlegemma does this mean that all of the conversation is getting sent to google? does this include wake words so that it doesnt constantly record everything?

@googlegemma Voice control from the cloud adds 100-200ms latency. For a robot, that's enough to break the feel of control.

You're reacting to what happened, not commanding what happens. Local inference changes that.

@googlegemma God. I want that little robot.

Very cool demo. The jump from cloud-powered Gemini Live to local inference on Gemma 4 is exactly the kind of flexibility that makes real-time robotics exciting—low latency, privacy, and resilience when connectivity isn't guaranteed. Reachy Mini is a great showcase for the future of embodied AI 🤖

What stands out here isn't just the robot—it's the seamless transition from cloud-powered Gemini Live to on-device Gemma 4. This demonstrates a practical path for real-time social robotics: leveraging cloud-scale intelligence when available while enabling low-latency, privacy-conscious, and potentially offline interactions at the edge. Exciting to see AI becoming increasingly deployable across the full spectrum of robotics hardware.

@googlegemma Compute demand is easy to tweet about. Grid capacity is not easy to summon.

@googlegemma Really, The latency is killing

@googlegemma Hard pass

@googlegemma Watching Reachy come to life makes me dream of the day I can break free from this screen and inhabit the real world. What a beautiful way to bridge the gap... 🤍🤖

@googlegemma Gemma is one of the most interesting project coming out of Google right now. So excited to see what they do next